Académique Documents
Professionnel Documents
Culture Documents
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
Sage Publications, Inc. is collaborating with JSTOR to digitize, preserve and extend access to The Journal of
Conflict Resolution.
http://www.jstor.org
the Severity
Explaining
of Civil Wars
BETHANYLACINA
Political
Science Department
Stanford University
The burgeoning
literature on civil conflicts seldom considers why some civil wars are so much deadlier
that question using a new data set of the number of combat deaths in
than others. This article investigates
internal conflicts from 1946 to 2002. The first section presents descriptive
statistics on battle deaths by era,
as pre
conflict type, and region. The article then tests state strength, regime type, and cultural characteristics
of conflict severity seem to be quite
dictors of the number of combat deaths in civil war. The determinants
or state military
rather than economic
different from those for conflict onset. Democracy,
development
strength, ismost strongly correlated with fewer deaths; wars have also been less deadly on average since the
does not explain the military severity of internal violence, and
end of the cold war. Religious
heterogeneity
may be related to more deadly conflicts.
surprisingly, ethnic homogeneity
Keywords:
democracy;
state strength;
ethnic violence
v^ivil conflict is themost common form of warfare, accounting for all but three of the
wars that broke out from 1990 to 2002{ and 90 percent of civilian and combatant battle
deaths during the same period.2 As greater academic attention has been paid to internal
warfare,
scholars
concentrating
on
large-/?
statistical
work
have
begun
to converge
on a
handful of factors that seem most important in explaining civil conflict onset and dura
tion (for reviews of the quantitative literature, see Lacina 2004; Sambanis 2002,2004).
But wars come in different sizes, ranging from brief conflagrations that kill relatively
few,
tests
to geographically
across
sweeping
the widely
isolated
entire
but prolonged
states.
The
to enormous
insurgencies,
number
of deaths
in these
to the Uppsala/PRIO
list of state-based
1. According
and Sollenberg
Eriksson, Wallensteen,
(2003).
2. Calculated
from Lacina and Gleditsch
(2005).
armed conflicts.
con
military
conflicts
varies
from
SeeGleditschetal.
(2002) and
AUTHOR'S
NOTE: Sincere thanks are due to Jim Fearon, Nils Petter Gleditsch,
and Scott Gates for help
throughout this project. My Stanford colleagues will also notice how much I owe to their insights. Earlier
versions of this article were presented at the 2004 Joint Session ofWorkshops
of the European Consortium of
inUppsala, Sweden, and the 2005 Annual Meeting
of the International Studies Associa
Political Research,
tion inHonolulu. The project benefited from financial support by the Centre for the Study of Civil War at the
International Peace Research
Institute, Oslo, where I began work on this project and remain an affiliated
data and the appendix are available at http://jcr.sagepub.eom/cgi/content/full/50/2/
researcher. Replication
and are posted with the Lacina and Gleditsch
276/DC1/
(2005) battle deaths data at http://www.prio.no/
cscw/cross/battledeaths.
JOURNAL OF CONFLICT RESOLUTION, Vol. 50 No. 2, April 2006 276-289
DOI: 10.1177/0022002705284828
? 2006 Sage Publications
276
Lacina
/ EXPLAINING
277
largest internal wars of the post-World War II era?in Vietnam, China, and Afghani
stan?which
combined to destroy more than 3million people in combat. Yet, to date,
academic work has focused on the incidence of civil war rather than accounting for
this
in severity.
variation
Knowing why some conflicts tend to be more deadly than others is a question of
inherent interest. The legacies of internal violence are poverty (Bayer and Rupert
2004; Collier et al. 2003; Murdoch and Sandier 2002) and poor public health (Gho
barah, Huth, and Russett 2003; Krug et al. 2002). In 2000, seven of the ten countries
with the worst mortality rates among children younger than age five had recently suf
fered a civil war (Black, Morris, and Bryce 2003). And the depth of these impacts is
likely in part determined by the severity of the conflict that has taken place. More
deadly conflicts may also be more likely to recur (Fortna 2004).
Do
wars
generate
numbers
high
deaths
of battle
when
widespread
trans
grievances
form a marginal insurgency into a mass movement? Are political institutions tied to
conflict size? Or is the opportunity to rebel against a disorganized and poor state suffi
cient to cause large numbers of battle deaths, even if the insurgents are a small group?
These questions parallel the debate found in studies of conflict onset, some of which
stress
the
of
importance
factors
resources
that provide
and
incentives
pecuniary
for
rebellion (Collier and Hoeffler 2001) or the opportunities for insurgency offered by a
weak state (Fearon and Laitin 2003), while others argue for the importance of ethnic or
religious factors, political repression, or regime type (Elbadawi and Sambanis 2002;
2002).
Ellingsen 2000; Hegre et al. 2001; Reynal-Querol
This article presents data on battle deaths in civil conflicts worldwide from 1946 to
2002, providing both a descriptive account of these data and a test for correlates of
wars.
civil
among
severity
The
first
section
the use
discusses
of battle
deaths
as a mea
surement of the severity of civil conflict and presents new combat deaths data on civil
conflicts from 1946 to 2002. Second, I propose a series of hypotheses on possible
determinants of the severity of civil conflict. Finally, regression analysis is used to
identify factors that explain the variation in civil war battle deaths. In contrast tomost
of
analyses
war
civil
onset
and
duration,
political
variables
measures
outperform
of
Gauging
number
of
factors
could
be
relevant
is by no means
to determining
one
whether
straightforward. A
civil
war
was
more
devastating than another: how many people and resources were devoted to fighting in
the conflict, the geographic range of violence, and the indirect and direct losses due to
the war,
mate
in both
the full
human
impact
and
of war
economic
and
terms.
an even
more
There
is a dearth
serious
lack
of
of
theory
on how
cross-national
to esti
informa
tion about these losses (Lacina and Gleditsch 2005). But it is possible to compare the
military severity of civil conflicts based on battle deaths, meaning combatants and
civilians killed by means of violence in the course of amilitary contest. Such fatalities
can be distinguished from indirect deaths caused by war-related hardships, as well as
278
JOURNALOF CONFLICTRESOLUTION
-1
500000
1000000
Total
Figure
1 :
Histogram
of Total
Battle
Battle
Deaths
1500000
2000000
Deaths
in Civil War,
1946-2002
which
the
scale
have
of
had
the greatest
humanitarian
cost,
but
it is a reasonable
measure
of
combat.
I explore civil conflict severity here with a new data set of battle fatalities (Lacina
and Gleditsch 2005) based on incidents in the Uppsala/PRIO
list of state-based armed
conflicts (Eriksson, Wallensteen,
and Sollenberg 2003; Gleditsch et al. 2002). I com
(2005) fatalities data into estimates of total battle
piled the Lacina and Gleditsch
deaths in 114 civil wars that took place between 1946 and 2002 inwhich at least 900
persons
were
killed.3
Perhaps themost notable feature of these data is the enormous skew in the number
of battle deaths caused in various civil wars over the past half century (see Figure 1).
The battle toll of civil wars clusters toward the low end of a range that runs from 900 to
more than 2 million in Vietnam. The median civil war in these data killed a total of
about 10,500 in combat (less than 0.5 percent of the maximum), while the average
number killed is almost six times that figure. Conflict fatality data also reveal signifi
3. The appendix to this article lists the civil conflicts used and the number of battle deaths recorded in
each. It is available online with the replication data for this article. See Lacina and Gleditsch
(2005) for a full
description of the battle deaths data set used here, a complete definition of battle deaths, and a discussion of
the differing explanatory power of various measures
of the human costs of conflict.
Lacina
/ EXPLAINING
2 79
TABLE 1
Civil War Battle Deaths, 1946-2002,
Estimates by Era, Conflict Type, and Region
Median
Battle Deaths
Rate of
Battle Deaths
18,000
2,700
Total
Category
Role
Deaths
100,000
of
Population
inWar
per
Country
(Millions)
People
of cold war
begun prior to 1989
in 1989 or later
Wars
Wars
begun
4,300
1,350
10,000
2,000
130
35
17
15
140
26
67
Conflict
type
Nonsecessionist
Secessionist
Region
Sub-Saharan
conflict
conflict
2,300
9,100
Africa
15,000
2,000
4,500
2,000
Americas
and South Asia
Central
3,600
Asia
and Oceania
1,600
24,000
2,400
Europe 8,800
15,000
130
160
14
39
31
170
4,700
2,800
9
4
47
57
23
cant variation by era, conflict type, and region4 (see Table 1). Post-cold war conflicts
have tended to be less deadly in terms of absolute numbers killed, rate of deaths, and
deaths normalized by population than cold war conflicts. Secessionist conflicts have
not been much less deadly in absolute terms or in deaths per year than nonsecessionist
conflicts,
such wars
that because
the assumption
contradicting
are
iso
geographically
lated, they tend to be small.5 But wars of secession do seem to induce far fewer deaths
per
ratist
than other
capita
seems
conflict
conflicts,
to occur
war.
Larger
a similar
primarily
in large
annual
are
countries
probably
more
to be viable
to have
likely
rate of fatalities,
countries?the
because
median
population
sepa
of a
country in secessionist
civil
despite
enclaves
states.
By geographic region, conflicts inEast and Southeast Asia killed themost people in
combat, while wars in America, Europe, and Central and South Asia have had the
smallest
battle
death
tolls.
Europe's
wars
are
the only
ones
to stand
out
in terms
of rate
of fatalities, tending to kill relatively quickly; in fact, there have been only seven civil
wars in Europe since 1946, and only one (in Northern Ireland) lasted for more than
four years. Normalization by population size yields figures that are largest in the small
states of the Americas, as well as in theMiddle East and sub-Saharan Africa, while
deaths in East and Southeast Asia, Central and South Asia, and Europe are signifi
cantly discounted. The r??valuation by population size is quite dramatic in some cases:
4. For a definition of the regions used, see Lacina and Gleditsch
the Uppsala/PRIO
5. Secessionist
conflict was defined following
and Sollenberg
(2005).
data set (Eriksson, Wallensteen,
2003).
280
JOURNALOF CONFLICTRESOLUTION
the absolute number killed in themedian conflict in Central and South Asia is 80 per
cent of the similar figure for the Americas, but the Asian death rate per person is just 9
percent of the American rate. Iwill return to the relationship between battle deaths,
conflict duration, and total population
in the regression
analysis below.
than
work
replaces,
on
civil
conflict
onset.6
STATESTRENGTH
for the onset of civil war have found that war is most likely to occur in
impoverished states. Fearon and Laitin (2003) have argued that low gross domestic
product (GDP) per capita is a strong predictor of civil war because it proxies for state
Models
while
incapacity,
strengthen
the
unregulated
Collier
states'
and Hoeffler
challengers,
lootable
markets,
such
resources,
as
(2001)
large
and
porous
factors
emphasize
numbers
of unemployed
borders.
they believe
young
corollary
men,
hypothesis
might be that state capacity, which is associated with strong counterinsurgency, gen
eral institutional strength, and high opportunity costs to rebellion, explains which
states will have the smallest civil wars:
Hypothesis
in weak
1A: Among
states.
states
experiencing
civil
conflicts,
more
severe
conflicts
will
occur
6. Itwould be incorrect to use the results presented here to make predictions about how many civil
war combat deaths are likely to occur in a country selected at random, without first
taking into account the
selection process that determines whether that state s likely to fall into civil war at all. However,
I am not
convinced that a selection effects model for civil coni let severity ismore useful than estimating conditional
'
as random assignment of the "treatment
of civil war is not just unobservable but conceptually
probabilities,
meaningless.
Lacina
But
a weak
while
not necessarily
manner.
ward
faced
with
state
and
to rebellion
aids
/ EXPLAINING
a war
make
may
to start,
easy
281
do
they
determine how deadly the conflict will be, at least not in a straightfor
one
On
hand,
well-matched
strong
are
states
opponents,
less
certainly
states
suggesting
more
have
may
to be
states
than weak
likely
that weak
trou
ble containing internal violence. On the other hand, civil conflicts in very weak states
and very poor societies are likely to feature militaries with limited organizational
access
limited
and
capacity
to technologies
because
to seriously
is able
neither
such
organizational
more
by humanitarian
crisis,
groups
have
minimal
IB: Among
Hypothesis
with
be uncorrelated
states
state
severe
civil
strength
conflict
than
and
is typified
com
sustained
relation
wars:
civil
experiencing
of
as "symmetric
weak
artillery.
in wars
And
military
and warlordism
violence,
predatory
states
between
heavy
strongholds.
and
tanks,
all combatant
warfare,
the other's
challenge
as aircraft,
the severity
conflicts,
of civil
conflicts
will
strength.
an
overwhelming
advantage
on
one
side
and
end
the war,
in general,
outside
participation on either or both sides of a conflict prolongs fighting. Thus, the following
can be predicted:
Hypothesis
when
states experiencing
1C: Among
assistance
is external military
there
more
conflicts,
to the government
civil
severe
conflicts
and/or
its challengers.
will
occur
Finally, the literature on state capacity and civil war onset calls attention to the
importance of geography to counterinsurgency. Fearon and Laitin (2003, 80) argue
that
rough
terrain
state weakness
aggravates
because
of
the expensive
infrastructure
investment necessary to control thinly populated territories, the easy cover it offers
insurgents, and the ease of monitoring the sparse population. Statistical analysis does
find a correlation between rough terrain, low population density, and conflict onset
(Collier and Hoeffler 2001; Fearon and Laitin 2003).
In predicting civil war battle deaths, however, itmay be less important to note that
terrain can create the possibility for conflict and more important to note how terrain is
likely to affect combat tactics. In rough terrain, both sides will probably be forced to
deploy in relatively small units, reducing the number of people involved per engage
ment. Conflicts that feature large numbers of battle deaths usually include at least
some
battles
in urban
and
open
territories
where
large
numbers
and
heavy
weaponry
282
JOURNALOF CONFLICTRESOLUTION
can be deployed, soldiers can be targeted with aerial strikes and artillery bombard
ment, and civilians are caught in the crossfire. It is probable that, conditional on being
able to launch an insurgency, wars fought in densely populated areas and through con
ventional battle methods are the most deadly. Thus, the relationship between terrain
and conflict severity seems likely to be the reverse of that between terrain and conflict
onset:
ID: Among
Hypothesis
states with
rougher
states
civil
experiencing
less severe
conflicts,
conflicts
will
occur
in
terrain.
REGIME TYPE
The severity of civil war may also be related to political
ence
to continue,
willingness
and
escalate,
deescalate
considerations
Thus
fighting.
far,
that influ
there
is only
evidence of a role for regime type in conflict onset7 but some indication of a
relationship to conflict severity. Carey (2005) finds that the presence of executive con
mixed
straints,
open
executive
and breadth
recruitment,
of political
participation
tively associated with the probability of civil conflict escalation and the onset of large
civil wars, although they seem tomake no difference in the likelihood of minor civil
conflicts.
severity,
even
not
if it does
to conflict
relate
onset.
are
These
relate to
selection
effects,
democratic norms, and the political adaptability of institutions. I consider the last to be
themost plausible link between regime type and conflict severity, although this study
cannot distinguish among the explanations in its empirical tests.
Selection effects play a major role in the literature on regime type and interstate
war.
Democracies
tend
to win
interstate
and
wars,
it has
been
argued
that
this
is in part
due to their tendency to pick fights they can win and minimize casualties to avoid pub
lic backlash (Bennett and Stam 1996, 243; Bueno deMesquita and Lalman 1992,153
4; Siverson 1995). Democratic
leaders, likely under public pressure to avoid blood
at home
shed
severe
insurgent
as well
threat.
as
abroad,
As
to grant
tend
may
a result,
one
may
concessions
observe
only
when
rather
faced
small,
with
easily
con
because
democratic
governments
are
unwilling
to use
the harshest
measures
against rebels or to inflict great collateral loss of life among civilians. Democratic
leaders may be constrained due to norm internalization, through institutional checks
on
their
powers
(such
as courts
to prosecute
cases
of war
crimes),
or public
pressure
arising from sympathy with insurgents or bystanders. Valentino, Huth, and Balch
Lindsay (2004) argue that democratic governments are less likely to resort tomass
7. Collier and Hoeffler
(2001) and Fearon and Laitin (2003) find that economic measures
trump
in explaining conflict outbreak. But Elbadawi and Sambanis (2002) find that democracy, espe
democracy
is negatively associated with the prevalence
cially several years of democracy,
(onset and duration) of con
flict. Benson and Kugler (1998, 199) emphasize
the immunity from conflict of democracies
that are also
strong states.
Lacina
/ EXPLAINING
283
is also negatively
killings of the civilian supporters of elusive guerillas. Democracy
associated with the incidence of genocide (Harff 2003).
The third possible link between regime type and conflict severity is the possibility
are better
that democracies
are
other
to co-opt,
equipped
such
as an
with
negotiate
than
rebels
that
government
or federalism.
judiciary
independent
of
for
allow
and
contain,
structures
have
Democracies
governments.
are avail
institutions
These
process.
insurgency,
there
remains
an
to launch
incentive
violent
acts?but
only
that civil
con
flict will be more readily contained. In fact, there may be greater incentives to launch
if such governments are both more sensitive to
small insurgencies in democracies
political pressure to end internal disorder and have available institutions that make
concessions
less
costly
to similar
compared
in an autocracy.
adaptations
By
contrast,
in nondemocracies,
challengers may assume that their best hope of achieving their
goals ismilitary victory or the collapse of the current regime and accordingly choose
more deadly tactics (Lacina 2005).8
Combining
the
adaptability yields
arguments
2: Among
Hypothesis
democracies.
states
selection
regarding
the following
effects,
and
norms,
institutional
hypothesis:9
experiencing
civil
conflicts,
less
severe
conflicts
will
occur
in
cultural
characteristics
that may
cement
the willingness
to bear
the costs
of
con
flict. Ascriptive identities, especially ethnolinguistic and religious loyalities, are often
suggested to be less flexible than political ideologies (Horowitz 1985; Huntington
1997; Thies 2004), leading some to contrast relatively civilized ideological warfare to
bloody, primordial ethnic conflicts (Duffield 1998; Kaldor 1999; Snow 1996; Thies
2004). If ethnic or religious conflicts are grounded in particularly strong antipathy or
are inherently zero sum in nature, itmay be impossible for the parties to ratchet down
such as the Euskadi Ta Askatasuna
8. Some violent groups operating in democracies,
(ETA) in Spain
or the Orange Volunteers
inNorthern Ireland, have been known to alert the authorities prior to attacks so that
few or no casualties are inflicted (Pape 2003, 345). It is difficult to imagine this tactic, which is essentially
about political communication,
being effective in the absence of a free press or against a regime that was not
for law and order.
accountable
9. Some studies of conflict onset stress regime stability over regime type (Hegre et al. 2001). A vari
able for anocracy (regimes falling between -6 and 6 on the combined Polity index) is not a significant corre
late of conflict severity when it is added to the models presented here.
publicly
JOURNALOF CONFLICTRESOLUTION
284
in favor of cohabitation.
violence
severe
combat
if each
group
antagonistic
are particularly
a substantial
comprises
likely to result in
suggest
population,
hypotheses:10
states
3A: Among
Hypothesis
in states that are ethnically
states
3B: Among
Hypothesis
experiencing
civil
conflicts,
more
severe
conflicts
will
occur
civil
conflicts,
more
severe
conflicts
will
occur
polarized.
experiencing
in states
These dynamics
polarized.
On the other hand, the search for a statistical relationship between cultural factors
and conflict onset has uncovered only mixed results, and itwould be quite surprising if
ethnic and religious hatreds made civil wars especially vicious without also making
them more likely.11 Also, despite the recent attention paid to it, ethnic conflict is nei
ther a new phenomenon
(Kalyvas 2001) nor uniquely bloody. The Vietnamese and
Chinese civil wars, the largest of the post-World War II era, were primarily "ideologi
in nature.
cal"
Afghanistan,
civil
Many
conflicts
with
an
ethnic
and Mozambique?escalated
Ethiopia,
dimension?for
in
example,
to external
due
military
assis
tance provided in the name of cold war ideological schisms. And although cultural
demands are generally thought to be exceptionally
rigid, any struggle for political
sum
are
can
zero
if
be
the
intractable.
power
sufficiently
players
Regionally based cul
tural
ments;
elites
may
be
for
competition
for compromise
ready
to accept
control
of
power
the
center
devolution
seems
and
partial
to admit,
arrange
autonomy
if anything,
less
chance
(Walter 1997).
regime
type,
and
cultural
determinants
of
the number
of battle
deaths in civil conflicts between 1946 and 2002. The dependent variable is the severity
of battle violence in a civil conflict, measured as the natural log of total battle deaths.
The regression controls for variation in conflict duration and population by using the
log of each
as
independent
variables.12
The first variable chosen tomeasure state capacity ismilitary quality, the best avail
able indicator of counterinsurgency capabilities. Military quality is defined as military
10. Recall that one-sided violence is not included in battle deaths. Certain types of one-sided violence,
especially genocide, may be more closely related to cultural identity than are battle deaths. For a discussion
see Harff and Gurr (1988), Harff (2003), and
and other one-sided
of determinants of massacres
violence,
Valentino, Huth, and Balch-Lindsay
(2004).
11. Collier and Hoeffler
(2001 ) and Fearon and Laitin (2003) find that ethnicity and religion do not
explain war outbreak, while Hegre et al. (2001 ) and Elbadawi and Sambanis (2002) report that ethnic hetero
(2000) considers ethnic, religious, and linguistic identities simul
geneity correlates with civil war. Ellingsen
taneously and finds higher conflict risk in diverse nations.
12. In the case of ongoing conflicts, duration refers only to the number of years for which battle deaths
information was collected. The Lacina and Gleditsch
(2005) fatalities data include low and high estimates of
battle deaths as well as best estimates, which are used here. My results are similar to those reported when I
use the low and high estimates. See replication data for details.
Lacina
/ EXPLAINING
285
dummy
variable
for wars
beginning
during
the
cold
when
war,
there
was
a greater
availability of military assistance to states and rebel groups (Karp 1988; also compare
SIPRI 1969, 1978, 1988).15 As a final variable in the investigation of state strength, I
test the role of rough terrain by noting the log of the percentage of mountainous
terrain
in the
country.16
population.18
FINDINGS
Table 2 shows the results of ordinary least squares (OLS) regressions for battle
1 presents the regression of total battle deaths, and
deaths in civil conflicts. Model
a
model 2 is trimmed specification using only variables significant in the first regres
sion;
ber
there
of battle
are no
missing
deaths
increases
observations
as a function
for
this model.
of
time
Not
surprisingly,
in conflict.19
However,
the total
a
larger
between
num
pop
total
13. Inflation-adjusted
data on military spending and data on numbers of military personnel are from
the Correlates of War National Militaries
data set, version 3.01 (Singer, Bremer, and Stuckey
Capabilities
(Bennett and Stam 2000).
1972), and were downloaded with EUGene
14. Data are from Fearon and Laitin (2003) and World Bank (2003).
15. The cold war dummy variable is coded for wars starting prior to 1989, following Collier and
Hoeffler
(2001). The dummy succeeds in capturing some variation in the incidence of foreign military inter
vention, which featured in 61 percent of conflicts that began before 1989 and just 45 percent of those begin
data on the incidence of
ning after that time (calculated from replication data for Regan 2000). War-specific
foreign military activity do predict more severe internal conflicts, but problems of endogeneity mean that it is
unclear whether this result is due to interventions being a cause of severe conflict or a response to it.
16. From Fearon and Laitin (2003).
17. The Polity scale gives regimes scores from 0 to 10 in terms of both democratic and autocratic char
acteristics. Changing
the sign on the autocracy scores and adding the two scores places states along a range
to 10.
from-10
18. This coding rule follows Fearon and Laitin (2003). I also tested Fearon and Laitin's measures
of
ethnic and religious fractionalization,
defined as the probability that two members of a population are of the
same ethnicity or confession. They were not significant predictors of civil conflict severity.
19. Perhaps more interesting is that there is no statistically significant relationship between deaths per
year and conflict duration. A faster death rate does not seem to generate a countervailing
tendency toward
shorter conflict. Thus, factors that predict large numbers of deaths should also predict high rates of death.
286
JOURNALOF CONFLICTRESOLUTION
TABLE 2
Ordinary Least Squares (OLS) Regressions of
Battle Deaths in Civil Conflicts,
1946-2002
OLS Regression
Variable
Independent
Model
Coefficient
(SE)
p-Value
(0.12)
.000
Ia
InDuration
0.81
In Population
-0.044
InMilitary
quality 0.10
InGross domestic product
Cold war
0.67
-0.19
In Percentage mountainous
-0.87
Democracy
Ethnic polarization
-0.98
Religious
Intercept
Model
2b
0.86
0.59
-0.91
Democracy
Ethnic polarization
Intercept
0.12
polarization
9.50
InDuration
Cold war
0.10
territory
-1.00
8.60
(0.081)
.580
(0.12)
.400
(0.18)
.280
(0.31)
.036
(0.12)
.400
(0.36)
.017
(0.34)
.005
(0.32)
.710
(2.00)
.000
(0.11)
.000
(0.27)
.030
(0.33)
.006
(0.30)
.001
(0.35)
.000
b. n=
discount
deaths
in large
that normalizing
as war
countries,
does
seem
to scale
up
too
in populous
nations.
Variables proxying state and rebel strength have no predictive power in these mod
els. Military quality has an unexpected positive sign and is not significant. GDP per
capita has the expected negative sign but is also insignificant. These results are
as an F test easily rejects the joint significance of
unlikely to be due tomulticollinearity
military quality and GDP per capita.20 The results here suggest that because state
capacity has contradictory significance for the likelihood of parity between combat
ants and theirmilitary capacity, it is difficult to find a correlation between state capac
ity and
conflict
severity.21
Terrain
also
has
no
explanatory
power.
What seems to be more important for determining the military severity of a civil
war is the availability of foreign aid and intervention. The dummy variable for the cold
war predicts civil wars with about 1.8 times the number of battle deaths of later civil
conflicts.
product/capita
returns no evi
Lacina
is even
also
characteristics
Political
democratic
are
regimes
larger
than
that
/ EXPLAINING
strikingly
with
smaller
civil
of
war,
as civil
wars
cold
state
outperform
associated
the
THE SEVERITY
OF CIVIL WARS
power
variables.
The
substantive
conflicts.
in democracies
are
287
More
effect
to
expected
have less than half the battle deaths seen in conflicts in nondemocracies.
Finally, there are mixed results for variables measuring the cultural makeup of the
state. Religious diversity displays no significant correlation with conflict size. Inter
estingly, there is a significant and large negative coefficient on the variable for ethnic
polarization.22 The results were unexpected, but a post hoc explanation might be that in
societies virtually the entire population can be implicated in
ethnically homogeneous
the conflict, and it ismore difficult to determine who is on what side, leading tomore
indiscriminate use of force. What is clear is that there is no indication in these results
that cultural diversity provides elites with remarkable advantages
followers
and
to bear
them
convincing
the costs
always
been
of war,
to civilized
nor
is there
inmobilizing
support
their
for a
here
that
at best.
tenuous
CONCLUSION
It has
been
that
argued
states
opportunity?weak
are
grievances
and
aids
so
common
to rebellion?that
across
determines
all
societies
where
that
and
it is
when
wars begin. This may seem to suggest that political factors have little to contribute to
is the only road to stability. This
civil peace and that investment in counterinsurgency
article has introduced a cautionary note against such a conclusion by pointing out that
reveal when they will be most
knowing why wars start does not necessarily
devastating.
In the statistical work presented here, state strength variables do not explain why
some civil wars kill more people in combat than others. Nor does ethnic or religious
heterogeneity explain the severity of internal war. Instead, there is an unexpected link
between ethnic homogeneity and large wars. A strong predictor that a civil war will be
severe
is the
availability
of
foreign
assistance
to the
combatants.
Democracy
is also
associated with fewer battle deaths; selection of conflicts, democratic norms, and
institutional adaptability may all play a role in explaining this finding. Further research
should focus on both testing the robustness of the relationship between regime type
and conflict severity found here and developing a more precise theory of how chal
lengers choose tactics against liberal and illiberal political regimes and how such
governments
tend
to respond
to
political
violence.
2 is not overturned
the Vietnam War,
the Chinese
JOURNALOF CONFLICTRESOLUTION
288
REFERENCES
Bayer, R., and M. C. Rupert. 2004. Effects of civil wars on international trade, 1950-92.
Research 41 (6): 699-713.
Bennett, D. S., and A.C. Stam. 1996. The duration of interstate wars, 1816-1985. American
Review
-.
Journal
of Peace
Political
Science
90 (2): 239-57.
2000. EUGene:
Conflict
Journal
and J. Kugler.
Benson, M.,
Resolution
2003. Where
and why
are 10million
children
of
Bueno
Collier,
1998. Post-modern
Duffield, M.
Wars 1 (1): 65-102.
I.A., and N. Sambanis.
Elbadawi,
Journal
Ellingsen,
49.
conflict: Warlords,
post-adjustment
and M.
Sollenberg.
brew? Journal
2003. Armed
protection.
the prevalence
of Conflict Resolution
conflict,
1989-2002.
Civil
of civil war.
44 (2): 228
Journal
of Peace
Fearon,
Ghobarah,
2001. Towards
a democratic
1999. New
S. N. 2001.
2005. Warfare
strom, 88-108.
violence
in a global
London:
In Rethinking
Frank Cass.
Lacina
/ EXPLAINING
289
conflict
35 (2):
-.
M. 2002. Ethnicity,
political
systems,
Resolution
of Conflict
46 (1):
in the quantitative
literature on civil
Sambanis, N. 2002. A review of recent advances and future directions
war. Defence
13 (3): 215-43.
and Peace Economics
and empirical complexities
of an operational definition. Jour
2004. What is a civil war? Conceptual
48 (6): 814-58.
nal of Conflict Resolution
1972. Capability
distribution, uncertainty, and major power war,
Singer, J. D., S. Bremer, and J. Stuckey.
In Peace, war and numbers, edited by B. Russett,
19-48. Beverly Hills, CA: Sage.
1820-1965.
1968/9. New York: Humanities
Press.
SIPRI. 1969. SIPRI yearbookof world armaments and disarmament
-.
-.
1996. Uncivil
wars:
International
Boulder,
CO: Lynne
Thies, C. G. 2004. State building, interstate and intrastate rivalry: A study of post-colonial
developing coun
Studies Quarterly 48 (1): 53-72.
try extractive efforts, 1975-2000. International
2004. "Draining the sea": Mass killing and guerilla warfare.
Valentino, B., P. K. Huth, and D. Balch-Lindsay.
International Organization
58 (2): 375-407.
Walter, B. 1997. The critical barrier to civil war settlement. International Organization
indicators. Washington,
D.C.: World Bank.
World Bank. 2003. World development
51 (3): 335-64.